Tool Calling and Plugins for Stem Cell Therapy

Data Characteristics in This Category

Stem cell therapy pharmacovigilance data primarily originates from clinical trial reports, real-world evidence (RWE) studies, patient-reported adverse events (AEs), and academic literature. Data update frequencies vary; clinical trial data typically updates regularly throughout the trial period, while patient reports and literature data continuously grow. Document structures for clinical trial reports generally follow ICH GCP guidelines, including detailed patient baseline information, treatment protocols, adverse event occurrences, severity, causality assessments, and outcomes. Real-world data can be more heterogeneous, encompassing electronic health records (EHR) and insurance claims data, with diverse fields and varying standardization levels. Adverse event descriptions often involve cell source (e.g., autologous, allogeneic), administration route, cell dose, manufacturing process, and patient-specific reactions. Units may include cell counts (e.g., 10^6 cells/kg), administration frequency, and duration.

Constraints Imposed by These Characteristics on Tool Calling and Plugins

The highly specialized and heterogeneous nature of stem cell therapy data places specific demands on tool calling and plugin configurations. For example, processing clinical trial reports requires plugins to parse structured and semi-structured text, extracting key entities such as CTCAE-coded adverse events and concomitant medications from the WHO-DD drug dictionary. Due to varying data update frequencies, tool calling strategies must be flexible; patient reports requiring high real-time processing may necessitate more frequent plugin calls. Heterogeneous data sources require tools with data standardization and mapping capabilities, such as unifying cell dose units from different sources. The complexity of stem cell therapy often leads to multifactorial adverse reactions. Plugins need to support complex logical judgments and multi-parameter correlated queries, for instance, querying liver injury events for a specific cell product when combined with immunosuppressants. For lengthy clinical reports, text segmentation strategies and summarization plugin configurations are crucial to avoid exceeding limits and losing information.

Configuration Guidelines

Configuration Item	Recommended Value	Rationale for Recommendation
`maxContext`	`32000` tokens	Stem cell therapy reports often contain extensive clinical details, requiring a longer context window to understand relationships.
`Chunk size` (Segment Length)	`800–1200` characters	Ensures each text block contains sufficient information while preventing individual segments from becoming too long and semantically fragmented.
`Recall count` (Recall Count)	`Top 8–12 entries` (Top 8–12 items)	Increases the likelihood of retrieving relevant adverse event patterns or similar cases from the knowledge base.
`Similarity threshold` (Similarity Threshold)	`0.78–0.85`	Balances recall and accuracy, filtering for highly relevant stem cell therapy adverse event information related to the query.
`PARSE_FILE_TIMEOUT_SECONDS`	`600` seconds	Provides ample parsing time when processing large clinical study reports or real-world data files.
`tool_code_timeout`	`120` seconds	Offers a longer execution time window for complex data processing or external API calls.

Common Pitfalls

HTTP 504 Gateway Timeout errors occur when calling external APIs. This happens because external services' response times exceed the preset tool_code_timeout when processing large-scale real-world data or performing complex bioinformatics analyses.
Plugin execution results in empty fields or incorrect formats. This may be due to inconsistent field naming or data types in stem cell therapy data sources, preventing the plugin's internal data parsing logic from matching correctly.
The model fails to correctly call configured tools for web searches, even with an external model integrated. This occurs when there are discrepancies in adaptation between the external model and FastGPT's tool calling interface, or the tool_description is unclear, preventing the model from accurately understanding the tool's purpose and parameters.

Verification Steps

Upload a clinical trial report containing typical stem cell therapy adverse events. Observe the knowledge base processing logs to confirm successful file parsing under the PARSE_FILE_TIMEOUT_SECONDS configuration and check if segment content is reasonable.
Construct a query targeting a specific stem cell product and adverse reaction, for example, "neurological adverse events of allogeneic mesenchymal stem cells for GVHD." Verify successful tool invocation and check if the recall count and similarity threshold filter relevant documents.
Design a process involving an external API call to simulate querying the latest safety alerts for a specific cell therapy drug. Check if the plugin executes successfully within tool_code_timeout and returns data in the expected format.
Test with a real-world dataset containing various fields and units to confirm the plugin correctly extracts and standardizes key information, such as cell doses like 1x10^7 cells/kg or 5e6 cells/kg.

The values provided are common starting points and should be measured against your own samples.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.